The finance sector is getting more comfortable when it comes to allowing machines to make decisions for themselves, and now the Financial Times reports (paywall) that JPMorgan will use machine learning to perform trades across all of its its global equities algorithms. Instead of relying on hand-coded rules developed by humans, the system, called LOXM, has learned from billions of past transactions how to buy and sell fast and, crucially, at the best price. Trials in Europe showed that it’s “more efficient” than existing systems (read, "it makes more money"). Clearly that’s compelling for JPMorgan, which will now roll out the software in Asia and America. But as we’ve pointed out before, using machine learning approaches to optimize financial transaction is a great idea—until something goes wrong. For now, AI lacks the transparency required to explain to customers why a particular decision was made. And that may not prove too popular when large sums of money are at stake.